Why a Repaired PDF Can Still Fail Archival Validation

Most real damage is structural, and structural damage is fixable

The great majority of real-world PDF damage, a file that refuses to open at all or opens with visible errors, comes down to a broken cross-reference table or a corrupted trailer, most often the result of an interrupted download or a failing disk partway through a save. That kind of damage is completely fixable. Rebuilding the file's internal structure, which is all Repair PDF has to do in that case, restores it to its original content with nothing lost, because the page content underneath was never touched in the first place. Only the index describing where to find it was broken. A reader that refuses the file is refusing to follow a broken map, not reporting that the pages themselves are missing.

Damage to the page content itself is a different problem

A damaged content stream on a specific page is a different kind of problem from a broken index. In that case the vector text and drawing instructions for that page are gone, not merely hard to locate, and no amount of structural repair can reconstruct data that no longer exists in the file. The practical response is to render whatever can still be recovered from that page as an image and run OCR PDF over it to recover readable text, while every other, unaffected page in the same document keeps its original vector content untouched throughout.

Being repairable and being PDF/A compliant are separate standards

A PDF that now opens correctly in every reader, with clean text and no visible errors anywhere, can still fail PDF/A archival validation entirely, which is the separate job PDF to PDF/A exists for. That is a different and considerably stricter standard, checking for different things: every font embedded in the file rather than merely referenced by name, colours tied to a declared, fixed profile, and no dependency anywhere on external resources or software the reader might not have available decades from now. Fixing a broken structure was never meant to guarantee any of that. Repair and archival compliance answer two different questions about the same file.

Why a validator's true failure is more useful than a false pass

Building PDF/A structure without relying on Ghostscript, which is deliberately avoided here because of its licensing terms, means assembling the required structure directly rather than using the standard conversion path most tools rely on. That makes it a best-effort process rather than a guaranteed one. Running every conversion through veraPDF, the reference validator that real archives and courts rely on, and showing its verdict rather than a reassuring checkmark, means a failure is reported because it is true. That is considerably more useful than a tool claiming a success it cannot back up.

Reading a specific failure is more useful than treating validation as pass or fail

A validation failure that lists the exact rules a document didn't satisfy, a particular font that wasn't embedded or a colour space that wasn't properly declared, is a useful, actionable result in a way that a bare pass-or-fail indicator never could be. Downloading the file regardless of the validation outcome and reading what failed means a document that is ninety-five percent of the way to full compliance doesn't get treated the same as one that has barely started. The failing rules point at what still needs fixing, rather than leaving the whole document to be approached again from scratch.

Choosing a repair strategy and a PDF/A level

Each tool asks one question first, and a wrong answer costs a run. Repair PDF has three strategies. Auto, the default, rebuilds the structure and falls back to an image plus OCR only on a page whose content stream is broken. Structure-only never does that: a broken page comes back blank, the better choice when every surviving page must keep real, selectable text. Rasterize-and-OCR redraws every page, readable ones included, as an image with recognised text, which suits a file whose words display as gibberish. The result screen names the pages recovered as images, so read it before passing the file on.

PDF to PDF/A asks for a level. 2b suits nearly anything made after 2010 and keeps transparency; 1b flattens any page with transparency or layers into a plain image, rarely what a modern file needs; 3b only adds the right to embed another file, such as an invoice's source spreadsheet. Two things matter more than the level: the standard forbids encryption, so a protected file needs Unlock PDF first, and a scan stays unsearchable unless the OCR option is on. The result is typically 10 to 40 percent larger: fonts that were not embedded are replaced with metrically compatible Liberation or Noto ones, and colours are bound to sRGB.